JP4950071B2 - Method for automatic extraction of pulmonary artery tree from 3D medical images - Google Patents

Abstract

An automated method (1) for the automatic extraction of a pulmonary vessel tree from a 3D medical image, such as multi-slice CT data, is disclosed. A segmented pulmonary vessel is identified as either an artery or a vein by determining a measure for arterialness for the vessel. The measure is based on a relation of the orientation of a local bronchus to the orientation of the segmented pulmonary vessel of the local bronchus. When a vessel is identified as a pulmonary artery, it is added to the pulmonary artery tree. Radii of the pulmonary artery and bronchus are measured automatically and positions where a ratio of these radii exhibits unusual values are presented in a display, preferably for suggesting further assessment by a radiologist, which for instance is useful for pulmonary embolism detection.

Description

  The present invention relates generally to the field of medical imaging. More specifically, the present invention relates to a method for automatically detecting a pulmonary artery tree from a three-dimensional medical image, for example, for detecting pulmonary embolism.

  Multi-row detector computed tomography (MDCT) allows acquisition of high resolution data of the entire chest. The quality of this data, together with the pulmonary artery tree, allows assessment of the bronchi. Extraction of the pulmonary artery tree is an important preprocessing step, for example for embolism detection. The quantitative assessment of the bronchial tree also includes important additional information that can be extracted from the MDCT data. Therefore, the simultaneous evaluation of the tracheobronchial tree and the accompanying pulmonary artery tree is indispensable for the diagnosis and treatment of asthma patients and emphysema patients. For example, the ratio of bronchial inner diameter to associated arterial diameter is an important parameter in clinical practice to detect and quantify airway stenosis and bronchodilation.

  When extracting a pulmonary vascular tree from multi-slice CT data, a problem arises in distinguishing the pulmonary artery from the vein. For example, a region extraction method based on a seed point for extracting a pulmonary artery has a problem of leakage into the pulmonary vein. The methods known so far extract all blood vessels, ie pulmonary arteries and veins, without making any distinction. Furthermore, the tracheobronchial tree and the pulmonary vascular tree are split separately, so the relationship between the tracheobronchial tree and the pulmonary vascular tree is not inherent in the split results. Therefore, these methods are not appropriate for clinical practice.

De Jong PA et al. (Pulmonary disease assessment in cystic fibrosis: comparison of values of bronchial and arterial sizing and CT assessment systems (comparison of CT scanning systems and value of measurement and measurement 1), Radiology 1) No. 2, pp. 434-439, May 2004) disclose a method for measuring the radius of the bronchi and associated arteries and for identifying bronchi / arterial pairs. However, the disclosed method is performed manually by a human observer. Therefore, this method is based on the mental activity of individual observers and as a result is prone to errors. For example, the quality of identification depends on the experience of the individual performing the extraction, the concentration of the individual at the time of evaluation, and the like.
Evaluation of pulmonary disease in cystic fibrosis: values of bronchial and arterial sizing and comparison of CT evaluation systems and values of bronchial and arterial dimension measurements, radiology, Vol. 231, no. 2, pp. 434-439, May 2004, De Jong PA, etc.

  Accordingly, it is an object of the present invention to provide an advantageous automatic performing arterial / venous separation in medical 3D images.

  Accordingly, the present invention preferably seeks to mitigate, mitigate or eliminate one or more of the above identified deficiencies and disadvantages in the art, one at a time or in combination, and in accordance with the appended claims. It solves at least the above problems by providing a computer readable medium and a medical examination apparatus.

  A general solution according to the method according to the invention comprises the step of automatically identifying the artery based on the associated bronchi.

  The method for arterial / venous separation is based on the insight that the pulmonary arteries run parallel to the bronchi, while the pulmonary veins tend to intersect perpendicularly with the adjacent airways as they approach the bronchi. The fact that the pulmonary artery passes parallel to the bronchi is used to divide the artery and remove the veins.

  If there are available bronchial trees and associated arterial trees, these bronchial trees and arterial trees may be used for joint visualization, for example, in virtual bronchoscopy applications. Bronchial and arterial diameters can be measured fully automatically along both the bronchial tree and the arterial tree during the splitting step. The location where the ratio of these radii shows an abnormal value is marked upon display and is presented for further evaluation by the radiologist.

  According to one aspect of the present invention, a method for automatically extracting a pulmonary artery tree from a three-dimensional medical image is provided. The method includes determining a location of a local bronchus, dividing a local lung vessel of the local bronchus into segmented local lung vessels of the local bronchus, and the segmented local lung of the local bronchus Determining the positioning of the blood vessels and dividing the segmented local pulmonary blood vessels into pulmonary arteries and pulmonary veins by determining the degree of arterialness of each segmented local pulmonary blood vessel.

  According to a further aspect of the invention, a computer readable medium embodying a computer program for automatic extraction of a pulmonary artery tree from a three-dimensional medical image is provided for processing by a computer. The computer program includes a code segment that determines the positioning of a local bronchus, a code segment that divides a local pulmonary blood vessel of the local bronchus into segmented local pulmonary blood vessels of the local bronchus, and the segmented of the local bronchus. A code segment that determines the positioning of the local pulmonary blood vessels and a code segment that divides the segmented local pulmonary blood vessels into pulmonary arteries and pulmonary veins by determining the degree of arterialness of each segmented local pulmonary blood vessel And have.

  In accordance with another aspect of the present invention, there is provided a medical examination apparatus arranged to perform the method according to the present invention, preferably a medical imaging workstation, configured to receive and process 3D images. The medical examination apparatus is configured for automatic extraction of a pulmonary artery tree from a three-dimensional medical image, and means for determining the positioning of the local bronchus, and local pulmonary blood vessels of the local bronchi are segmented into the local bronchi By means of dividing the segmented local pulmonary vessels, means for determining the positioning of the segmented local pulmonary vessels of the local bronchus, and determining the degree of arterialness of each segmented local pulmonary vessel, Means for dividing the segmented local pulmonary blood vessels into pulmonary arteries and pulmonary veins.

  The present invention is useful compared to the prior art in that it advantageously provides suppression of the pulmonary veins from the pulmonary vascular tree, in applications and visualization modes relating solely to the pulmonary artery tree.

  These and other aspects, features and advantages that may be realized by the present invention will be apparent from and will be elucidated with reference to the accompanying drawings, from the following detailed description of embodiments of the invention.

  The following description focuses on embodiments of the present invention applicable to computed tomography (CT) systems, particularly multi-slice CT data. However, it will be appreciated that the present invention is not limited to such applications and many other imaging including, for example, magnetic resonance imaging (MRI) systems, three-dimensional ultrasound (3D-US) systems, and the like. Can be applied to the system.

  The following describes an automated method for the extraction of a pulmonary artery tree from multi-slice CT data utilizing a tracheobronchial tree. If there are available bronchial trees and associated arterial trees, these bronchial trees and arterial trees may be used for joint visualization, for example, in virtual bronchoscopy applications. Bronchial and arterial diameters are measured completely automatically along both bronchial and arterial trees. The locations where these ratios of radii exhibit anomalous values are marked for display and are presented for further evaluation by the radiologist (see, eg, white dots / marks in FIG. 4).

  In a first embodiment of the method of the present invention, with reference to FIG.

  Initially, in step 10, the bronchial tree is transformed into, for example, Schlaterter et al. )) Is extracted using a suitable algorithm such as based on the algorithm described in

  Then, that is, if bronchial tree splitting is applicable from step 10, the localization of the local bronchi is estimated at step 11. This can be done, for example, based on the centerline orientation of the bronchi, or alternatively, by using an indicator based on, for example, a gray-value. This is described in detail below.

  Then, at step 12, segmentation based on pulmonary vessel seed points is performed. During this time, the positioning of these pulmonary vessels is estimated locally. This is described in more detail below.

  For each of the blood vessel segments extracted in step 12, a magnitude based on the absolute value of the inner product of the blood vessel segment and the nearest bronchus orientation vector is estimated. This size has a high value for blood vessels passing in parallel near the bronchi. This size makes it possible to determine whether the currently considered blood vessel is a pulmonary artery or a pulmonary vein. That is, it is considered the degree of “arterialness” of the blood vessel segment. The magnitude value is used to determine whether the vessel segment is included in the arterial tree. This may be done by comparison with a threshold value. Preferably, the threshold is determined empirically. An alternative method includes integrating a value indicative of arterialness across the branch including the parent and child segments of a given segment.

  If at step 12 the vessel is identified as belonging to the arterial lung tree, at step 13 the radius of the vessel is determined. The local positioning measurement described above works as follows. That is, when considering the center line point of the blood vessel, the average radius deviation of the image data is calculated in a sphere having the center at this center line point. A large negative deviation occurs due to a sudden change in gray value when the sphere touches the blood vessel boundary. The radius of the sphere varies over a predetermined range, and in this case, the radius where the average deviation is minimum is considered an estimate for the radius of the blood vessel.

  Then, in step 14, the vessel positioning is estimated from the radius deviation values in the optimal radius sphere. For each given pair of antipodal points in the sphere, the sum of absolute radius deviations is calculated. The sum for each pair is considered to represent the position of the blood vessel, assuming that the change in gray value is minimal along the blood vessel direction. The vector connecting this pair of antipodal points is a positioning estimate.

  The same procedure may also be used for the bronchi if the gray value profile is reversed in this case, and the maximum mean radius deviation is searched rather than the minimum value in radius estimation. .

  If there are available bronchial trees and associated arterial trees, these bronchial trees and arterial trees may be used for joint visualization, for example, in virtual bronchoscopy applications. Bronchial and arterial diameters are measured fully automatically along both the bronchial tree and the arterial tree during the segmentation step described above, according to the method. The locations where the ratio of these radii shows an abnormal value are marked for display and may be presented for further evaluation by the radiologist.

  A further embodiment 130 of the method of the present invention is represented in FIGS. In the previous example, a pre-segmented bronchial tree was considered. When classifying a blood vessel into an artery and a vein, the centerline point of the existing bronchial tree is used as a candidate point. In the following examples, an alternative approach is used. Here, when examining a blood vessel in detail, a candidate point for a bronchus is searched based on a gray value distribution nearby. This method can give better results under certain conditions. This is because the bronchial tree extraction misses the branch and therefore all subtrees of the bronchial tree are not found, which can cause the corresponding arterial subtree to be incorrectly labeled as “vein”. is there. This is avoided in the embodiments described in more detail below by measuring vessel and bronchi positioning in a more robust manner.

  The present method, similar to the previous method for arterial / venous separation, is based on the fact that the pulmonary artery tree is associated with the bronchial tree. For each extracted blood vessel segment, the degree of “arterialness” is determined. This current degree combines two factors: a method for identifying candidate locations with respect to the bronchi passing near a given vessel, and a repositioning indicator for vessel segments and bronchial candidates. The latter element provides a blood vessel that runs parallel to the nearby bronchi. Spatial positioning of vessel segments and bronchi is estimated by applying a structural tensor near the local gray value.

  In experiments conducted to perform the method, a lung multi-slice CT dataset was used. These data sets were acquired by a Philips IDT 16-slice and a Philips Brilliance 40-slice scanner. Experimental results have shown that the measurement incorrectly reduces the number of pulmonary veins contained in the arterial tree.

  The method provides an automated method for extracting a pulmonary vascular tree from a multi-slice computed tomography (MSCT) dataset, and further provides a method for distinguishing pulmonary arteries from veins.

  The tree splitting method begins with one or more seed positions. From this position, front propagation is started. It can be seen that even if the seed point is placed in the pulmonary artery, various sub-trees are also considered to be included in the split. This is due to the fact that multiple intersections of arteries and veins appear with the same Hounsfield value and the data set does not show separation between vessels. Such arterial / venous intersections can easily be mistaken for arterial tree bifurcations.

  The method for distinguishing pulmonary arteries from pulmonary veins is based on the fact that the arterial tree is associated with the airway tree. After segmentation of the vascular tree, each vascular segment (our terminology, meaning the part of the vascular tree that starts at one branch point and ends at the (far)) next branch point is close to it. Checked for bronchi. When bronchi are found, the vessel and bronchi positioning are compared. The closer the structure is to parallel, the higher the “arterialness” value assigned to this vessel.

  The following sections describe the vascular tree splitting method. More specifically, lung division is represented by FIGS. 3A-3F show the different steps of lung segmentation in an example axial slice. 3A and 3B show the upsampled results of region expansion applied to the subsampled data set. 3A shows upsampling for a complete 3 × 3 × 3 block, and FIG. 3B shows upsampling according to the HU threshold. FIG. 3C represents the division after removal of the thick airway. FIG. 3D represents the division after closing the gap (closing) to include pulmonary vessels, where the closing is applied to the entire lung mask. It should be noted that a “bridge” appears to connect the left and right lungs at the arterial portion of the lung, as in the circled region in FIG. 3D. FIG. 3E represents the result when the closing operation is applied separately to the left and right lungs to avoid unwanted bridging effects. FIG. 3F gives the final segmentation result, and FIG. 4 is the maximum intensity projection of that final result.

  Here, this division will be described in more detail.

  Before actual vessel tree extraction is performed, the lungs are segmented from the data set in a preprocessing step. This reduces the search space for vessel tree segmentation. First, the trachea is automatically detected. For this purpose, techniques for detection of the descending aorta in the MSCT data set are used. The work is performed on a data set in a subsampled form. A subsampling rate (subsampling rate) of 5 in the x, y and z directions is used. A preferred threshold of −600 HU is applied, and the “black area” (Hounsfield value <−600 HU) has an x-direction range between 10 and 30 mm and a y-direction range between 10 and 50 mm. Be detected. The candidate points obtained in this way are clustered and the centroid of the point cluster with the most expressed positioning in the z direction is selected as the most promising trachea seed candidate. As a result, see FIGS. This procedure found the trachea shown by the cross sections in FIGS. 2A-2C accurately in all of the data sets tested so far.

  Intermediate results of the lung segmentation procedure to be described below are shown in FIGS. A simple region expansion process starting from the tracheal seed point is started, including all connected voxels that fall below a given Hounsfield value of -600 HU. Region expansion is performed on the subsampled data set, where the subsampling rate is 3 in all directions. The result of region expansion is reflected in all data sets by mapping each contained voxel to the corresponding 3 × 3 × 3 block. Since including the entire 3 × 3 × 3 block corresponding to the segmented voxel can result in the inclusion of unwanted bright voxels (see FIG. 3A), the same threshold used for region expansion is upsampled. Applied in steps. Thus, within a given block, only voxels below −600 HU are included (see FIG. 3B). Before the blood vessels are included in the previously reached lung segmentation, the trachea and main bronchi need to be removed to avoid filling in the space between the lungs and trachea and blood vessels. The trachea is divided using the same region expansion procedure and the seed points described above. This time a preferred threshold of -900 HU is used. The tracheal mask is expanded by 3 mm and then subtracted from the lung mask (see FIG. 3C). The expansion is done via a fast marching algorithm that expands the outer wall surface of the tracheal mask used as the initial front. Here, a procedure based on a two-dimensional slice is used for tracheal segmentation.

  In the next step, blood vessels not included in the previously extracted lung mask are filled. This is done by a line based procedure. The row is removed from the lung mask and the gap is searched. A gap having a length shorter than a predetermined maximum length is closed. According to the implementation, first all the y-direction lines are processed with a maximum length of dy = 15 mm, and subsequently the x-direction lines are processed with a maximum length of dx = 15 mm. Finally, the z-direction line is processed with dz = 5 mm. If the entire lung mask is processed in this way, one is at risk of including a bridge between the left and right lungs (see FIG. 3D). To avoid this, the left and right lungs are processed separately (see FIG. 3E). The final result shown in FIGS. 3F and 4 is obtained by removal of bright areas at the surface of the mask included during the vessel filling step.

  The methodology used for local vessel / bronchi positioning measurements is described in the next section.

  The procedure consists of three levels: voxel level, segment level, and tree level.

  [Voxel level]: High-speed marching and front propagation processing starting from the initial seed point is started. The voxel acceptance criteria determines whether the voxel is being touched. This may be, for example, a HU threshold criterion. Propagated fronts are confirmed for connectivity at regular intervals. This is considered a branch in the tree when the front is divided into two or more disconnected elements. In this case, the currently expanded segment is complete and one new segment for each connected front element is initialized. In order to keep the almost flat front part in shape, the front propagation speed is kept constant ν≡1.

  [Segment level]: During the segment expansion, it is confirmed by the termination criterion whether the expansion of the current segment should be continued or terminated. Typical termination causes are front splits or reaching the maximum allowable segment length. Segment evaluation criteria are evaluated on a regular basis and have the potential to reject a segment. With respect to the accepted segment, a new segment is initialized for each connected front element. On the other hand, the rejected segment does not initialize its children. The segment evaluation criterion includes a criterion related to a geometric characteristic such as a maximum segment radius and an elongated shape, or a criterion based on gray values such as a tubular property and an orientedness. Segments that are ultimately unacceptable can be replayed using various criteria of auto-adapted parameter sets. These parameters can also be adapted whenever a new segment is inherited from a parent segment.

  Tree level: At this level, decisions are made that cannot be found in a single segment. For example, a subtree is removed if the density of branch points becomes too high to indicate a leak.

  Along with the division, a tree structure including a branch point and a blood vessel center line are formed. The center line point is estimated as the center of weight of the propagated front element. The described process is represented in FIGS.

In Figure 5, the tree split is:
Seed point 50
Center line point 51
Junction 52
Connecting front element 53
Segment 54
Represented by

In FIG. 6, the following methods for constructing the tree structure are:
Seed point 60
Initialize 1 segment for each seed point 61
Add segment to segment column 62
Tree removal 63
No column? 64
Yes 65
No 66
Extract first segment from column 67
Extended and continuous assessment 68
Evaluation Results 69
Expansion continuation 70
Parameter conformity 71
Reset segment and adapt extended parameters 72
Segment exclusion 73
Segment acceptance 74
Initialize one new segment for each connected CFC element 75
Tree removal 76
Completion 77
Represented by

In this example, at the voxel level, all voxels having a Hounsfield value of -600 HU or above are accepted. The expansion is terminated at the bifurcation point, i.e. when the propagating wave front is separated. At the segment level, a segment is identified for its radius. If the radius exceeds the maximum allowed value, the segment is rejected. The maximum allowable radius r _max of a segment is determined by the actual radius of its parent segment r _parent , where r _max = 2r _parent . At the tree level, end segments with a length shorter than 5 mm are removed. The result of tree splitting is shown in FIGS. 7A and 7B. The result of segmentation includes a set of voxels, a centerline tree containing branch and parent-child information, and vessel radius estimates for each centerline point.

  [Positioning estimation]: In order to calculate the degree of arteriality described above in step 135, the positioning of local blood vessels and bronchi is determined together with directivity in steps 131-134 (see FIG. 13). The directivity indicates how much the linear position of the local gray value structure can be identified. A structural tensor is used to obtain an estimate of these values.

として定義される。 The structure tensor J is

Is defined as

と表される。 Where g _x , g _y , and g _z are three-dimensional image gradient vectors,

It is expressed.

のガウス重み関数が用いられる。 The parentheses <·> in the formula (1) indicate a weighted average by the weight function w. In the experiment,

The Gaussian weight function is used.

Let e ₀ , e _1, and e ₂ be the eigenvectors of the structure tensor J sorted so that the corresponding eigenvalues are in ascending order λ ₀ <λ ₁ <λ ₂ . Thus, e ₀ corresponds to the positioning with the smallest absolute direction deviation, while e ₂ corresponds to the positioning with the largest deviation. When calculated inside a blood vessel or bronchus, the minimum gray value deviation is detected along the tubular structure, so that e ₀ is used as a local positioning estimate.

は、局部濃淡値近傍の構造化（ｓｔｒｕｃｔｕｒｅｄｎｅｓｓ）の程度である。平面的な濃淡値構造で、２つの小さな、理想的にはゼロである固有値が求められる。従って、最大及び２番目に最大の固有値の正規化された比率：

は、濃淡値構造の平面化（ｐｌａｎｅｎｓｓ）を示す。留意すべきは、Ｓ≧Ｐである点である。差分：

は、局部濃淡値構造の線形な位置付けの程度として用いられる。位置付け推定結果の例及び対応する指向性表示値（ｏｒｉｅｎｔｅｄｎｅｓｓ ｖａｌｕｅ）は図８に与えられる。図８で、肺血管（ａ）及び２つの気管支（（ｂ）及び（ｃ））の直交断面が示される。これらは、上述されたように、構造テンソルの固有ベクトルｅ_０、ｅ_１及びｅ_２に従って方向を合わせられる。表される例に関する指向性表示値は、（ａ）０．９１、（ｂ）０．８７、及び（ｃ）０．５１である。 Within the tubular structure, i.e. inside the blood vessel or bronchus, one position with a low gray value deviation and two orthogonal positions with a larger deviation are detected. Normalized difference between maximum and minimum eigenvalues:

Is the degree of structuredness near the local gray value. In a planar gray value structure, two small, ideally zero eigenvalues are found. Thus, the normalized ratio of the largest and second largest eigenvalues:

Indicates the planarization of the gray value structure. It should be noted that S ≧ P. Difference:

Is used as the degree of linear positioning of the local gray value structure. An example of a positioning estimation result and the corresponding directional display value are given in FIG. In FIG. 8, an orthogonal cross section of a pulmonary blood vessel (a) and two bronchi ((b) and (c)) is shown. These are oriented according to the eigenvectors e ₀ , e ₁ and e ₂ of the structure tensor, as described above. The directivity display values for the depicted example are (a) 0.91, (b) 0.87, and (c) 0.51.

  Here, the degree of “arterialness” will be outlined.

  In order to distinguish an artery from a vein, the degree of arterialness is determined. According to this embodiment of the method according to the invention, this degree of arterialness gives a value, for example between 0 and 1, along the blood vessel centerline for each sample point. A value of 0 indicates that the centerline point is likely to be part of a vein, while a value of 1 indicates that the point being viewed is most likely to be part of an artery. The main step in estimating the arterialness of the blood vessel centerline point is the search for bronchial candidate points that are considered to be associated with the artery.

[Detection of homologous bronchial candidates]: Detection is started by estimating the above-mentioned vascular positioning as an initial candidate group for several given vessel centerline points. Let r _ves be the local blood vessel radius at the centerline point. Σ = r _ves / 2 is used for the Gaussian weight function in equation (3). The surface normal to vessel positioning is resampled trilinearly. A two-dimensional square image I composed of 64 × 64 pixels, which is an example covering an 8r _ves × 8r _ves region, is extracted by a blood vessel center line point at the center thereof. The two-dimensional section is smoothed by a Gaussian filter with σ = r _ves / 4. Some examples are shown in FIG. FIG. 9 shows cross sections of two different example blood vessels. Five cross sections are shown in 1 mm increments along each of the two example vessels. Candidate bronchial candidates in the reduced candidate group are marked with a white cross. The five cross-sections at the top of FIG. 9 show an artery with a radius of 3 mm. Note that the candidate group does not contain a false component in this case. As can be seen, homologous bronchi are correctly identified. The estimated arterialness of this segment is 0.98. The five cross sections at the bottom of FIG. 9 are veins with a radius of 3 mm. It should be noted that there are a large number of candidate points for each of the slices. This is due to the fact that in the reduction step from the initial candidate group to the reduced abnormal value group, there is no bronchi with a current high Hessen value that can be an “outlier”. The arterialness for this segment is 0.28.

  All local gray value minimum positions Pi within the range of the smoothed cross-sectional image I are considered as possible bronchial candidate points and constitute an initial candidate group.

の行列式が計算される。この値は、例えば、誤った浅い局部最小値に関してよりも、実際の気管支候補に関してより大きい。仮に、
（外１）

を、所与の血管中心線点に関する全ての初期気管支候補の中の最大ヘッセン値とする。Ｈｉ＜Ｈ_ｍａｘ／２を有する全ての候補点Ｐｉは、この場合に、初期候補群から取り除かれる。更に、Ｉ（Ｐｉ）＞−８００ＨＵであるところの全ての候補点Ｐｉが捨てられる。残りの候補は、低減された候補群を構成する。 Here, a reduced candidate group is considered. In the following, the local minimum is removed and the initial candidate group is reduced. For each candidate point, the Hessian matrix:

Is calculated. This value is greater for an actual bronchial candidate than for an incorrect shallow local minimum, for example. what if,
(Outside 1)

Is the maximum Hessen value among all initial bronchial candidates for a given vessel centerline point. All candidate points Pi with Hi <H _max / 2 are removed from the initial candidate group in this case. Furthermore, all candidate points Pi where I (Pi)> − 800HU are discarded. The remaining candidates constitute a reduced candidate group.

[Degree of arterialness]
The reduced candidate group is used to estimate the arterial likelihood of the blood vessel at a given centerline point. First of all, it should be noted that if there are a large number of points in the reduced candidate group, there is a high probability of looking at a vein rather than an artery. This is because, in the case of arteries, the Hessian value Hi for the correct bronchus is usually much higher than for other local minimums. Therefore, the correct candidate is an abnormal value in the distribution of Hessian values. For example, if all values below half of the maximum value are discarded, this will yield most false candidates. Thus, in the case of arteries, most false candidates are discarded when going from the initial candidate group to the reduced candidate group. On the other hand, in the case of veins, such an abnormal value does not exist, and a large number of local minimum values can exist as candidates even in the reduced candidate group. For this reason, arteriality is included in the reduced candidate group concentration, ie, the group, in order not to obtain a large value indicative of arterialness from one of many current candidate points. A centerline point is assigned a value of 0 if the number of points is greater than a certain number. In the experiment, an example value of 5 was selected.

と評価される。 Alternatively, for candidate Pi included in the reduced candidate group, the mutual positioning Ci between the blood vessel and the pseudobronchi is:

It is evaluated.

The positioning e _pi and directivity Oi at the candidate point Pi are estimated by a structure tensor having σ = r _ves / 4 with respect to the Gaussian weight function used in the calculation of the structure tensor. If the directivity Oi at the candidate point Pi is smaller than the minimum directivity Omin, the positioning estimation is extremely unreliable, and the possibility of the presence of bronchi is low. Thus, the arterialness is set to zero in this case. In an exemplary implementation, Omin is set to 0.2.

のように定義される。 The arterial character of the centerline point is:
(Outside 2)

Is defined as follows.

  The arterialness of the segment, i.e. the part of the blood vessel between the two bifurcation points, is defined as the average value of the values representing the arteriality of up to 50% of the centerline point. Excluding 50% of the points ensures robustness to unreliable values, for example, close to the bifurcation point.

  The method 130 provided was applied to several MSCT data sets (Phillips Brilliance 40 slices). For evaluation of the method, large blood vessels leaving the heart and entering the heart were excluded. This is because in these parts, arteries and veins tend to have crossings that pass in close proximity to each other and do not allow them to be split as separate blood vessels using a gray value-based approach. It is. The analysis was applied to blood vessels with a radius of 7 mm or less. Arteries with a radius between 3 mm and 7 mm were reliably identified. In this radius range, the artery undergoes a very high arterialness rating. A blood vessel was classified as an artery when the arterial likelihood was 0.8 or greater by comparison with the threshold. If the arterialness is less than 0.8, the blood vessel is considered a vein.

  The risk of falsely marking smaller arterial branches as veins increases with decreasing vessel radius. This is because small blood vessels as bright tubular structures are more visible than smaller sized bronchi. The small bronchial wall is subject to a partial volume that limits its detectability. Furthermore, the bronchial radius is much smaller than the radius of the associated artery. For this reason, the application of arterialness was limited to blood vessels with a radius of at least 3 mm. According to experience gained from experiments, branching structures below this vessel size are unlikely to contain error-prone arterial-venous intersections. Thus, unmarked subtrees of marked vessel segments received the same label as their parent segment. The effect that blood vessels can be seen to the end compared to the bronchi is shown in FIG. In FIG. 10, the vessel tree segmentation result is overlaid with the airway tree segmentation using a method similar to the method presented by Schlaterter et al. The rendering of the image of FIG. 10 is based on centerline tree information including radius values for each centerline point. FIG. 10 shows a tracheobronchial tree with blood vessels classified into pulmonary arteries and pulmonary veins.

  All arteries associated with the bronchi that are accurately found by the airway segmentation used are marked correctly.

  In summary, this example of the method according to the invention has been described. A method for extracting a pulmonary vascular tree from an MSCT data set and a method for assigning arterial values to all vascular segments of the segmentation results have been described. Such an arterial value is estimated for blood vessels within a certain radius range. Of the segments with very large vessel radii, many correspond to two or more vessels that pass close to each other or intersect each other, rather than the actual anatomical vessel portion. For very small blood vessels, homologous bronchi are not visible at a given resolution. Since the presented method relies on bronchial visibility, such small blood vessels are excluded from direct analysis. In the middle range of vessel radius, the presented approach works robustly. The arterial / venous indicia is propagated from the intermediate vessel to the terminal sub-tree because the smaller portion of the vessel sub-tree is considered to have fewer vessel crossings.

  In the experiment, the presented method accurately identifies most arteries. The method provides a useful step in the use of the bronchial tree for arterial / venous separation.

  In another embodiment according to FIG. 11, a computer readable medium is schematically represented. The computer readable medium embodies itself a computer program 110 for automatic extraction of the pulmonary artery tree 111 from the three-dimensional medical image 112 for processing by the computer 113. The computer program 110 includes a code segment 114 that determines the positioning of the local bronchus, a code segment 115 that divides the local pulmonary blood vessels of the local bronchus into segmented local pulmonary blood vessels of the local bronchus, and a segmented local of the local bronchi A code segment 116 that determines the positioning of the pulmonary vessels, and a code segment 117 that separates the segmented local pulmonary vessels into pulmonary arteries and pulmonary veins by determining the degree of arterialness of each segmented local pulmonary vessel Have

  FIG. 12 represents an exemplary medical workstation in accordance with a further embodiment of the present invention. A medical workstation is arranged to perform the method of the present invention and is configured to receive and process a three-dimensional medical image. Preferably, the workstation is arranged to execute the program code segment described above to perform the method according to the invention. According to this embodiment, the medical workstation 119 is configured for automatic extraction of a pulmonary artery tree from a three-dimensional medical image, and means 120 for determining the positioning of the local bronchus, and the local pulmonary vessels of the local bronchus as a segment of the local bronchus Means 121 for dividing the segmented local pulmonary vessels, means 122 for determining the location of the segmented local pulmonary vessels of the local bronchus, and determining the degree of arterialness of each segmented local pulmonary vessel And means 123 for dividing the segmented local pulmonary blood vessels into pulmonary arteries and pulmonary veins. These means 120-123 are preferably electrical components that are operatively connected to each other in a suitable manner. Other parts of the medical workstation, for example as a display or pointer actuator, are not depicted or described in detail.

  As a result, automatic extraction of a pulmonary vascular tree from a three-dimensional medical image such as multi-slice CT data is disclosed. Segmented pulmonary vessels are identified as either arteries or veins by determining the degree of arterialness of the vessel. This degree of arteriality is based on the relationship of local bronchi positioning to local bronchi segmented pulmonary vessel positioning. When a blood vessel is identified as a pulmonary artery, it is added to the pulmonary artery tree. The pulmonary artery and bronchi radii are automatically measured, and the location where the ratio of these radii shows an abnormal value is preferably presented on display for further evaluation by the radiologist. This is useful, for example, for pulmonary embolism detection. An example is given in FIG. In FIG. 4, the white dots are marks of such positions in the pulmonary artery tree. FIG. 4 also shows the silhouettes of all lobes surrounding the displayed pulmonary artery. This allows the radiologist or other medical personnel to easily find the location of interest.

  The application and use of the above method according to the present invention varies, for example to a CT scanner controller, imaging workstation, or Picture Archiving Communication System (PACS) workstation as an option, such as a software option. Includes example application. The method can be easily introduced into an existing system in terms of calculation, and is flexible.

  Application of the method can be easily detected since the results are usually displayed visually. According to one embodiment, for example, the display shows the connection of the tracheobronchial and pulmonary arterial tree with a display of potential affected areas (see FIG. 10).

  The invention can be implemented in any suitable form including hardware, software, firmware or any combination of these. However, preferably, the invention is implemented as computer software running on one or more data processing devices and / or digital signal processing devices. The elements and components of an embodiment of the invention may be physically, functionally and logically implemented in any suitable way. In practice, the function may be implemented in a single unit, in multiple units, or as part of another functional unit. In this way, the present invention may be implemented in a single unit or may be physically and functionally distributed between different units and processing devices.

  Although the present invention has been described with reference to particular embodiments, it is not intended to be limited to the particular forms recited individually. Rather, the present invention is limited only by the accompanying claims, and other embodiments than the specific embodiment described above, for example, trachea locations, seed points for certain algorithms other than those described above, Various methods for obtaining a method for determining the tracheobronchial tree and the like are also possible within the scope of these appended claims.

  In particular, it is indicated to determine whether a blood vessel is an artery or a vein based on the nearby bronchi rather than extracting the entire pulmonary vascular tree. Conversely, it is possible to extract the bronchial tree and search for accompanying blood vessels that can be marked as “arteries” in the vicinity of each bronchus, but blood vessels are not even extracted at all by such operations. However, although it seems that such a method is possible, it is not as optimal as the method described above because bronchial tree extraction is more difficult than blood vessel extraction and some bronchi can be easily missed. As some bronchi are missed, the corresponding arteries can also be missed. This is undesirable.

  In the claims, the word “comprising” does not exclude the presence of other elements or steps. Moreover, even if individually listed, a plurality of means, elements or method steps may be implemented by eg a single unit or processing unit. Further, although individual features may be included in different claims, these may be advantageously combined in some cases, and inclusion in different claims may indicate that a combination of features is feasible and / or not advantageous. Not implied. Moreover, only one reference does not exclude a plurality. The words “one”, “first”, “second”, etc., do not exclude a plurality. Reference signs in the claims are provided merely as a clarifying example shall not be construed as limiting the scope of the claims in any way.

4 is a flowchart representing an embodiment of a method according to the present invention. A to C are CT slice images as examples showing the results of the trachea type discovery procedure. A cross-thoracic slice (512 × 512) is displayed, and the white cross marks the estimated tracheal position. A to E are example CT slice images representing the lung segmentation method, and F is a CT image representing the result of the method represented by A to E. 4 is a CT image representing the results of the method represented by FIGS. It is a schematic diagram showing a tree splitting algorithm. It is a flowchart of a tree division algorithm. It is the maximum intensity projection (MIP) of the extracted vascular tree. FIG. 6 is a rendering of an extracted vascular tree based on centerline information including only a radius. It is a figure which shows the orthogonal cross section of a pulmonary blood vessel. 2 is a portion of a CT slice showing a cross section of two blood vessels. Figure 3 is a three-dimensional image representing a segmented and labeled pulmonary vascular tree overlaid with an airway tree. 1 is a schematic diagram of an embodiment of a computer readable medium of the present invention. FIG. FIG. 6 is a schematic diagram of an exemplary medical workstation in accordance with a further embodiment of the present invention. 6 is a flowchart representing another embodiment of a method according to the present invention.

Claims (12)

A method for automatic extraction of a pulmonary artery tree from a 3D medical image comprising:
Determining the positioning of the local bronchi,
Dividing the local pulmonary blood vessel of the local bronchus into segmented local pulmonary blood vessels of the local bronchus;
Determining the location of the segmented local pulmonary blood vessel of the local bronchus, and determining the degree of arterial character of each segmented local pulmonary blood vessel, thereby determining the segmented local pulmonary blood vessel as a pulmonary artery And dividing into pulmonary veins,
Having a method.   The degree of arteriality of the segmented local pulmonary blood vessels is based on the relationship of the positioning of the local bronchus to the positioning of the segmented local pulmonary blood vessels, and the degree is determined as the local pulmonary artery of the local bronchi The method of claim 1, wherein a standard value is provided for the blood vessel to identify a blood vessel that passes in parallel near the local bronchus. The degree of arterialness of the segmented local pulmonary blood vessel is the absolute value of the inner product of the segmented local pulmonary blood vessel and the nearest local bronchus orientation vector;
The method of claim 2, wherein the degree provides a standard value for the blood vessel to identify a blood vessel that passes in parallel near the nearest local bronchus as a local pulmonary artery of the local bronchus. The standard value is given by thresholding the value of the degree of arterialness to determine whether the segmented local pulmonary vessels are included in the pulmonary artery tree;
The method according to claim 2 or 3, wherein an empirically derived threshold is used for the thresholding.   If the segmented local pulmonary vessel is determined to be a pulmonary artery in the step of dividing the segmented local pulmonary vessel into a pulmonary artery and a pulmonary vein, the segmented local pulmonary vessel has a radius and a given pulmonary artery The method of claim 1, further comprising: determining a radius of the pulmonary artery as a radius of a minimum average deviation of image data calculated in a sphere centered at a centerline point. The method of claim 1, wherein dividing the local pulmonary blood vessel of the local bronchus is based on seed points. The method of claim 1, wherein determining the local bronchi positioning is based on the local bronchus centerline positioning.   The method of claim 1, wherein determining the positioning of the local bronchus is based on using an indicator based on local intensity values. Measuring the diameter or radius of the local bronchus and the pulmonary artery along both the bronchial tree and the accompanying pulmonary artery tree;
And marked in the position that exhibit an abnormal value in the local bronchus, and the step of calculating the measured diameter or radius of the ratio of the pulmonary artery, and the calculated ratio is displayed, the abnormal position by the user Steps suggesting further evaluation,
Have
The measuring, calculating and marking steps determine local bronchial positioning, segment local bronchial pulmonary blood vessels, determine segmented local pulmonary blood vessel positioning; 9. The method of any one of claims 1 to 8, wherein the method is automatically performed in parallel with the step of dividing and dividing the segmented local pulmonary blood vessel into pulmonary arteries and pulmonary veins. A computer-readable medium embodying a computer program for automatic extraction of a pulmonary artery tree from a three-dimensional medical image for processing by a computer:
A code segment that determines the positioning of the local bronchi,
A code segment that divides the local pulmonary blood vessel of the local bronchus into segmented local pulmonary blood vessels of the local bronchus;
The segmented local pulmonary vessels are determined by determining a code segment that determines the location of the segmented local pulmonary vessels of the local bronchus, and the degree of arterialness of each segmented local pulmonary vessel. A code segment that separates the pulmonary artery and vein,
A computer-readable medium having:   The computer program according to claim 10, capable of executing the method according to claim 1. 10. A medical examination apparatus arranged to perform a method according to any one of claims 1 to 9, which is a medical image workstation configured to receive and process a three-dimensional image:
Means for determining the positioning of the local bronchi,
Means for dividing the local pulmonary blood vessel of the local bronchus into segmented local pulmonary blood vessels of the local bronchus;
Means for determining the positioning of the segmented local pulmonary blood vessels of the local bronchus, and determining the degree of arterialness of each segmented local pulmonary blood vessel, thereby determining the segmented local pulmonary blood vessels to the pulmonary artery And means for dividing into pulmonary veins,
Have
A medical examination apparatus configured for automatic extraction of a pulmonary artery tree from a three-dimensional medical image.

Images